SOX-mandated Internal Control Deficiency Disclosure under Section 302 and Earnings Quality: Evidence from Cross-listed Firms

Transcription

1 SOX-mandated Internal Control Deficiency Disclosure under Section 302 and Earnings Quality: Evidence from Cross-listed Firms Guojin Gong Smeal College of Business Pennsylvania State University Bin Ke Smeal College of Business Pennsylvania State University Yong Yu Red McCombs School of Business University of Texas at Austin December 1, 2008 We thank Jere Francis, Bill Kinney, Steve Kachelmeier, Jay Lee, Andy Leone, Thomas Lys, Sarah McVay, Karl Muller, May Zhang, and workshop participants at the Seoul National University, Binghamton University, University of Texas at Austin, University of Missouri-Columbia, Northwestern University, Penn State University, Singapore Management University, the 2008 International Symposium on Auditing Research, and the 18 th Annual Conference on Financial Economics and Accounting for helpful comments and Michael Crawley and Walid Al-Issa for able research assistance. Part of this research was conducted while Bin Ke visited the City University of Hong Kong. Corresponding author Bin Ke Pennsylvania State University 316 Business Building University Park, PA 16802, USA (phone) (fax) ( ) Electronic copy available at:

2 SOX-mandated Internal Control Deficiency Disclosure under Section 302 and Earnings Quality: Evidence from Cross-listed Firms Abstract We examine the usefulness of SOX-mandated internal control deficiency (ICD) disclosure under Section 302 in assessing earnings quality for cross-listed firms relative to U.S. firms. Consistent with prior research, we find that U.S. firms Section 302 ICD disclosure conveys useful information about earnings quality. However, cross-listed firms Section 302 ICD disclosure is on average unrelated to earnings quality and significantly less informative about earnings quality than U.S. firms. We provide evidence that the reduced usefulness of cross-listed firms ICD disclosure is due to management's weaker incentive to detect and report existing ICDs. Specifically, the weaker association between the ICD disclosure and earnings quality for crosslisted firms relative to U.S. firms is primarily driven by cross-listed firms domiciled in weak investor protection countries. In addition, cross-listed firms propensity to disclose ICDs declines with the degree of management s private control benefits and this effect is stronger for firms domiciled in weak investor protection countries. Key words: Internal control deficiency; Earnings quality; Cross-listed firms; Sarbanes-Oxley Act. Electronic copy available at:

3 I. Introduction The Sarbanes-Oxley Act of 2002 (SOX) is probably the most far-reaching U.S. securities law since the passage of the Securities and Exchange Act of 1933 and Among SOX's sweeping reforms of corporate governance, auditing, and financial reporting of publicly traded companies, the internal control provisions under Sections 302 and 404 are regarded as the most significant and also the most costly provisions of the SOX. An important requirement of Section 302 is that managers must evaluate the effectiveness of their firms internal control over financial reporting and disclose their conclusions on the internal control effectiveness and any material changes in internal control since the last periodic financial report. Section 404 further requires company management to include a separate internal control report in the annual filings with the SEC and requires the external auditor to attest to management s assessment of firms internal control effectiveness. Section 302 was effective immediately after the passage of SOX for both U.S. firms and cross-listed firms while Section 404 was phased in gradually over time starting with accelerated U.S. firms in November, It is a well accepted belief among researchers and practitioners (see, e.g., PCAOB, 2004) that a material weakness in internal control system implies more than a remote likelihood that a material misstatement of the financial statements will not be prevented or detected. Thus, if company management has exerted reasonable effort to detect and truthfully disclose existing ICDs, the ICD disclosure made under either Section 302 or Section 404 should be informative about earnings quality. Consistent with this expectation, recent research finds that U.S. firms Section 302 ICD disclosure provides useful information about earnings quality (see, e.g., Ashbaugh-Skaife et al., 2008a; Doyle et al., 2007a). Ashbaugh-Skaife et al. (2008a) further show that U.S. firms that report remediation of previously reported ICDs confirmed by their external 1

4 auditors under Section 404 exhibit an improvement in earnings quality, while those firms that fail to remediate ICDs continue to exhibit poorer earnings quality relative to non-icd disclosing firms. The objective of this study is to examine whether the Section 302 ICD disclosure made by foreign firms traded on the U.S. stock exchanges (referred to as cross-listed firms) is as informative about earnings quality as the Section 302 ICD disclosure made by U.S. firms. Following the recent SOX-mandated ICD disclosure literature (i.e., Ashbaugh-Skaife et al., 2008a; Doyle et al., 2007a), we use Dechow and Dichev s (2002) earnings quality proxy to assess the association between the Section 302 ICD disclosure and earnings quality. We also conduct sensitivity checks using alternative earnings quality measures, including the magnitude of abnormal accruals estimated from the modified Jones (1991) model, the magnitude of performance-matched abnormal accruals following Kothari et al. (2005), earnings management proxies used in Leuz et al. (2003), and the persistence of earnings with respect to future earnings or future cash flows. Recognizing that these accounting-based earnings quality proxies are subject to measurement errors, we also use the earnings response coefficient (ERC), a marketbased earnings quality proxy, to assess the informativeness of the Section 302 ICD disclosure with regard to earnings quality. Specifically, we examine the difference in the ERC between ICD disclosing firms and non-icd disclosing firms for cross-listed firms versus U.S. firms. Over the past decade, cross-listed firms have become an increasingly important component of the U.S. equity market. From 1990 to 2000, the number of foreign firms listed on NYSE and NASDAQ grew steadily from about 170 to over 750 with cumulative trading volume in these firms reaching more than $750 billion. As of 2002, foreign firms listed on NYSE represented nearly 17% of all NYSE listings (Coffee, 2002). In light of the rising importance of 2

5 cross-listed firms to U.S. financial markets and the importance of internal controls to financial reporting, we believe that it is important to study the informativeness of cross-listed firms ICD disclosure concerning earnings quality. More importantly, the existing evidence on the informativeness of U.S. firms ICD disclosure about earnings quality may not apply to cross-listed firms because the Section 302 ICD disclosure is not subject to the attestation of the external auditor and managers of crosslisted firms have weaker incentive than managers of U.S. firms to establish a sound internal control system and to expend resources and effort in detecting and truthfully disclosing ICDs. First, many cross-listed firms are domiciled in countries where investor protections are weak and management possesses substantial private control benefits as evidenced by the divergence between the management s voting rights and cash flow rights. The private control benefits create a strong incentive for managers to expropriate minority shareholders while weak home country investor protections reduce managers costs of such expropriation (see La Porta et al., 1998; Lins, 2003). A deficient internal control system further enhances managers ability to expropriate minority shareholders. As the disclosure of ICDs forces managers to implement remediation procedures to eliminate the ICDs, 1 which in turn reduces managers opportunity to exploit minority shareholders, managers of cross-listed firms likely have weaker incentive to detect and report ICDs than managers of U.S. firms. Second, the extant literature finds that cross-listed firms face lower risks of the SEC enforcement and U.S. shareholder litigation than U.S. firms, which would reinforce cross-listed firms incentive to conceal existing ICDs. As argued in Licht (2003), the U.S. regulatory regime that applies to foreign firms is significantly inferior to that faced by U.S. firms and the SEC has 1 As indicated in SEC staff s response to Frequently Asked Questions, if the registrant were to identify a material weakness, it should carefully consider whether that fact should be disclosed, as well as changes made in response to the material weakness (italics added) (SEC Release No , #9). 3

6 largely adopted a hands-off enforcement policy toward cross-listed firms. Siegel (2005) shows that for the entire period since the enactment of the 1933 Securities and Exchange Act, there are virtually no reports regarding public enforcement steps for cross-listed firms, even when egregious misconduct has been publicized in cross-listed firms home countries. Siegel also finds little evidence of minority shareholders success in pursuing litigation cases against cross-listed firms who violated U.S. securities laws. Of course, there also exist countervailing forces that may motivate managers of crosslisted firms to exert effort to detect and truthfully report existing ICDs. The SOX was enacted in the aftermath of several high-profile accounting scandals and the U.S. stock market crash. Due to significant changes in the financial reporting and regulatory environment after the passage of the SOX, corporate boards, investors, and the SEC could become more vigilant about potential corporate governance issues and scrutinize firm management s internal control disclosures more carefully. Indeed, Coffee (2002) finds that the SEC has recently brought high-profile enforcement actions against cross-listed firms in the U.S. 2 It is ultimately an empirical question whether cross-listed firms Section 302 ICD disclosure is as informative about earnings quality as U.S. firms Section 302 ICD disclosure. Consistent with prior research, we find that U.S. firms that disclose ICDs have lower earnings quality (measured using either the accounting-based earnings quality proxies or the earnings response coefficient) than U.S. firms that do not disclose any ICDs. In sharp contrast, we find that the relation between the ICD disclosure and earnings quality is significantly weaker for cross-listed firms than for U.S. firms. In fact, we find no evidence that cross-listed firms that disclose ICDs have lower earnings quality than cross-listed firms that do not disclose any ICDs. 2 Although non-u.s. firms with deficient internal control systems could delist from U.S. stock exchanges (e.g., by shifting listing location to OTC or other countries such as United Kingdom) to avoid compliance with the SOX, the delisting requirements are quite restrictive and costly (see, e.g., Hostak et al., 2007; Marosi and Massoud, 2008). 4

7 These results suggest that cross-listed firms Section 302 ICD disclosure is not useful to separate high quality earnings from low quality earnings. We perform several analyses to understand the causes for the lack of informativeness of cross-listed firms Section 302 ICD disclosure regarding earnings quality. First, we show that cross-listed firms weaker results are consistent with managers lack of incentive to detect and disclose ICDs as discussed earlier. Specifically, we find that the weaker results for cross-listed firms are primarily driven by cross-listed firms domiciled in weak investor protection countries. In addition, cross-listed firms whose managers possess significant private control benefits are less likely than cross-listed firms whose managers do not possess significant private control benefits to disclose a Section 302 ICD, especially for cross-listed firms domiciled in weak investor protection countries. Second, we perform a battery of robustness tests to rule out the possibility that the weaker results for cross-listed firms are due to inherent limitations of our data sources or research design. Specifically, we determine that the weaker results for cross-listed firms are not caused by (1) the difference in accounting standards used by cross-listed firms and U.S. firms, (2) the overreporting of ICDs by cross-listed firms relative to U.S. firms, and (3) the difference in the relation between ICDs and earnings quality for U.S. firms and cross-listed firms even in the absence of managerial incentive to conceal existing ICDs (see Section 6 for the details). We make two contributions to the literatures. First, we contribute to the emerging literature examining the information content of SOX-mandated ICD disclosure. The extant literature focuses on the ICD disclosure made by U.S. firms. The major issues examined in this literature include the economic determinants of the ICD disclosure (Ashbaugh-Skaife et al., 2007; Doyle et al., 2007b), the usefulness of the ICD disclosure in assessing earnings quality 5

8 (Ashbaugh-Skaife et al., 2008a; Doyle et al., 2007a) and cost of capital (Ghosh and Lubberink, 2006; Beneish et al., 2008; Ashbaugh-Skaife et al., 2008b; Ogneva et al., 2007), and the stock market reaction to the disclosure of ICDs (Beneish et al., 2008; Hammersley et al., 2007). To our knowledge, we are the first to examine cross-listed firms ICD disclosure and its implications for earnings quality. Our findings should be of interest to investors and regulators. For investors who wish to use the SOX-mandated ICD disclosure to assess cross-listed firms earnings quality, our evidence suggests that they should interpret with caution the financial reports prepared by crosslisted firms that disclose no Section 302 ICDs because there is no evidence that these firms earnings quality is higher than that of cross-listed firms that report Section 302 ICDs. For regulators who are evaluating the benefits of SOX, our evidence raises concerns over the effectiveness of cross-listed firms implementation of Section 302 due to their managers incentive to conceal existing ICDs. Second, we contribute to the literature on cross-listed firms financial reporting behavior, especially managerial disclosure without the involvement of the external auditor. While there is ample empirical evidence suggesting that cross-listed firms financial reporting quality is superior to that of their domestic counterparts (see Karolyi, 2006), empirical evidence is relatively scarce on the difference in financial reporting quality between cross-listed firms and U.S. firms. An exception is Lang et al. (2006) who find that cross-listed firms earnings quality is on average lower than that of U.S. firms. We contribute to this literature by showing why the Section 302 ICD disclosure, which is not subject to the attestation of the external auditor, serves as a reliable indicator of U.S. firms earnings quality but is not useful to separate high quality 6

9 earnings from low quality earnings for cross-listed firms. 3 Our results highlight the limitations of cross listing as a bonding mechanism and suggest that cross listing alone is unlikely to be a complete substitute for building home-country investor protection institutions. The rest of the study is organized as follows. The next section discusses the internal control disclosure requirements of the SOX. Section 3 describes the sample and data. Section 4 discusses the two primary regression models we use to test the association between the Section 302 ICD disclosure and earnings quality. Section 5 presents the empirical results on the association between the Section 302 ICD disclosure and earnings quality across cross-listed firms and U.S. firms. Section 6 analyzes possible reasons underlying the weaker association between the Section 302 ICD disclosure and earnings quality for cross-listed firms relative to U.S. firms. Section 7 reports additional sensitivity checks using alternative earnings quality proxies. Section 8 concludes. 2. Institutional background on SOX-mandated internal control deficiency disclosures Signed into law on July 30, 2002, the Sarbanes Oxley Act mandates a series of changes in corporate governance and financial reporting for public companies that are listed on major U.S. stock exchanges (see Zhang, 2007 and Coates, 2007 for detailed discussions of the SOX provisions). A major component of the Act is the internal control disclosure requirements under Sections 302 and 404. Section 302 requires that company management certifies the accuracy of the periodic financial reports filed with the SEC, evaluates the effectiveness of the firm s internal control over financial reporting, and discloses their conclusion on the internal control 3 In this study we focus on the ICD disclosure under the Section 302 reporting regime rather than the ICD disclosure under the Section 404 reporting regime because the former affords us a better opportunity to identify the managerial disclosure incentives that would be suppressed in the presence of the external auditor. In a future study we plan to examine whether the involvement of the external auditor under Section 404 will improve the informativeness of cross-listed firms ICD disclosure about earnings quality. 7

10 effectiveness and any material changes in internal control since the last periodic financial report. Section 404 further requires company management to include an explicit and separate internal control report in the annual filings with the SEC, and requires the external auditor to attest to management s assessment of firms internal control effectiveness. All SEC filers (both U.S. firms and cross-listed firms) are required to comply with the Section 302 disclosure requirements for fiscal years ending on or after August 29, Section 404 s effective date varies across U.S. firms and cross-listed firms. For cross-listed (U.S.) accelerated filers, Section 404 became effective for fiscal years ending on or after July 15, 2006 (November 15, 2004). 5 For both U.S. and cross-listed non-accelerated filers, Section 404 becomes effective for fiscal years ending on or after December 15, Cross-listed firms that trade on the over-the-counter market or issue private equity under SEC Rule 144a are exempt from the SOX. In this study we focus on the internal control disclosure made under the Section 302 reporting regime which is not attested by independent auditors. Internal control deficiencies can be generally classified into three types in the order of ascending severity: deficiency, significant deficiency, and material weakness. According to Auditing Standard (AS) No. 2 issued in March 2004, a control deficiency exists when the design or operation of a control does not allow management or employees, in the normal course of performing their assigned functions, to prevent or detect misstatements on a timely basis (AS No. 2, paragraph 8). A significant deficiency is a control deficiency, or combination of control deficiencies, that adversely affects the company s ability to initiate, authorize, record, process, or 4 Brown et al. (2007) report that Germany is the only foreign country that adopted internal control regulation prior to No German firms cross listed on U.S. stock exchanges have reported any Section 302 ICD since the enactment of the SOX. Results are similar if we exclude German firms from our non-icd disclosure sample. 5 An accelerated filer refers to a company that (1) has a public float of at least $75 million, (2) has been subject to the SEC's periodic reporting requirements for at least 12 months and has filed one annual report, and (3) is not eligible to use the SEC's small business reporting forms. 8

11 report external financial data reliably in accordance with generally accepted accounting principles such that there is more than a remote likelihood that a misstatement of the company s annual or interim financial statement that is more than inconsequential will not be prevented or detected (AS No. 2, paragraph 9). A material weakness is a significant deficiency, or combination of significant deficiencies, that results in more than a remote likelihood that a material misstatement of the annual or interim financial statements will not be prevented or detected (AS No. 2, paragraph 10). 6 The above categorization by AS No. 2 implies that material weaknesses should have the most severe consequences on financial misreporting than deficiencies or significant deficiencies, and thus should have the strongest relation with poor earnings quality. However, since AS No. 2 was issued well after many firms issued their first ICD disclosure, firms might have used different thresholds when reporting their ICD types (Ashbaugh-Skaife et al., 2007). Therefore, we conduct our empirical analyses using disclosures of material weaknesses as well as disclosures of all three types of ICDs. As the empirical results are similar using either material weakness disclosures or all types of ICD disclosures, we report our regression results for material weakness disclosures only for brevity. 3. Sample selection procedures and data sources 6 The SEC Releases No (Amendments to rules regarding management s report on internal control over financial reporting) dated June 20, 2007 and the SEC Releases No (Definition of the term significant deficiency) dated August 3, 2007 further clarified the definitions of a material weakness and a significant deficiency. A material weakness is redefined as a deficiency, or a combination of deficiencies, in internal control over financial reporting such that there is a reasonable possibility that a material misstatement of the registrant s annual or interim financial statements will not be prevented or detected on a timely basis. A significant deficiency is redefined as a deficiency, or a combination of deficiencies, in internal control over financial reporting that is less severe than a material weakness, yet important enough to merit attention by those responsible for oversight of the registrant s financial reporting. 9

12 We restrict our sample to the U.S. firms and cross-listed firms that are listed on the three major U.S. stock exchanges (NYSE, AMEX, or NASDAQ) since foreign firms that trade overthe-counter or issue private equity under SEC Rule 144a are exempt from the SOX. Consistent with the existing cross-listing literature (see, e.g., Karolyi, 1998, 2006; Reese and Weisbach, 2002), our cross-listed firm sample includes both American Depository Receipts (ADRs) and foreign firms directly listed on the U.S. stock exchanges (e.g., many Israeli and Canadian firms). Regardless of the form of U.S. listing, all foreign firms are incorporated in their home countries and thus are subject to similar home country institutional forces (e.g., company law and enforcement). Our sample of cross-listed firms is identified based on Compustat s variable FINC, which indicates a foreign firm s country of incorporation, and cross-checked with other data sources such as SEC filings, CRSP, and Audit Analytics. Firms that disclosed ICDs are identified based on Audit Analytics Disclosure Controls database. This database covers all SEC registrants who have disclosed management s certification of internal controls under Section 302 of the SOX in periodic SEC filings (including 10-K, 10KSB, 10-Q, 10QSB, 20-F and 40-F) since September We further require the availability of firm characteristics included in our regression models from Compustat (see models (3) and (4) described in Section 4). With these restrictions we generate a sample of 438 unique cross-listed firms, of which 80 firms disclosed at least one Section 302 ICD (material weakness, significant deficiency, or deficiency) and 49 firms disclosed at least one material weakness in internal controls during the period from September 2002 to July 2006 (the effective date of Section 404 for cross-listed accelerated filers). 7 There are 358 cross-listed firms that never disclosed any ICD during this 7 The disclosed ICDs by cross-listed firms relate to ineffective control environment, inadequate qualified staff who are familiar with U.S. GAAP, complexity of transactions such as derivatives, taxes, and stock option compensation, etc. Due to the small sample size, we do not separately analyze each category of ICDs in empirical analyses. 10

13 time period. 8 The foreign countries that have at least 10 unique firms included in our sample are Canada (90 firms), Israel (53), United Kingdom (47), Bermuda (19), Japan (22), Netherlands (21), France (14), Mexico (15), British Virgin Islands (12), and Chile (11). Because Canada and U.K. are generally considered as having stronger investor protection than the other foreign countries (see Aggarwal et al., 2007) and firms from these two countries comprise a large portion of our sample, we also perform all of our regression analyses after excluding Canadian and U.K. firms and obtain qualitatively similar results (untabulated). As noted in Section 2, accelerated U.S. filers ICD disclosures made after November 15, 2004 are subject to both Sections 302 and 404 reporting requirements. Because Section 404 requires auditor attestation to management s assessment of internal control systems, the characteristics of the ICDs disclosed before and after Section 404 s effective date may not be comparable (e.g., Doyle et al., 2007a). To facilitate the comparison with the Section 302 ICD disclosures by cross-listed firms, we exclude accelerated U.S. filers ICD disclosures made after November 15, 2004 (the effective date of Section 404 for U.S. accelerated filers). We identify U.S. accelerated filers as firms incorporated in the U.S. with market value of equity greater than $75 million at the end of the most recent second fiscal quarter prior to November 15, Our U.S. firm sample contains 3,332 unique firms, of which 524 firms reported at least one ICD 8 To ensure the accuracy and completeness of the Audit Analytics database, we read all 20-Fs or 40-Fs filed through Edgar by cross-listed firms in our sample during the period September 2002-July The details of internal control problems are provided in the Item Controls and Procedures in 20-Fs or various places in 40-Fs. Although each registrant is required to report whether their disclosure controls are effective or ineffective, most registrants use qualifying languages in their filings such as reasonably effective, effective, however, effective, subject to, effective, however our auditors have disclosed material weaknesses. Despite these non-standard reporting terms, the classification of ICD disclosures by Audit Analytics appears reasonably accurate and the coverage of the database seems complete. 11

14 (material weakness, significant deficiency, or deficiency) and 327 firms reported at least one material weakness under the Section 302 reporting regime. 9 As Section 404 s effective date is earlier for accelerated U.S. firms than for cross-listed firms, there is a concern that the differential informativeness of the Section 302 ICD disclosure across U.S. firms and cross-listed firms is driven by the accelerated U.S. firms who could be under greater pressure to detect and truthfully report existing ICDs during the Section 302 reporting regime in anticipation of the imminent external audit under Section 404. We find that all of our regression analyses are robust to deleting the accelerated U.S. firms, suggesting that the timing difference in Section 404 s effective dates for U.S. firms and cross-listed firms is not a driver of our results. 4. Research design We employ two complementary approaches to examine the usefulness of the Section 302 ICD disclosure in assessing earnings quality. The first approach examines the association between the Section 302 ICD disclosure and Dechow and Dichev s (2002) earnings quality measure (referred to as the accounting-based approach). Our second approach uses the earnings response coefficient (ERC) to infer earnings quality (referred to as the market-based approach). The two approaches have different strengths and weaknesses and thus are complementary to each other. The advantage of the accounting-based approach is that it directly measures earnings quality from reported accounting information, but a well recognized disadvantage is that accounting-based measures of earnings quality are noisy and even biased. The primary 9 Since U.S. firms may report ICDs in form 8-Ks but Audit Analytics does not collect ICDs disclosed in form 8-Ks, we cross-checked the sample of U.S. firms ICD disclosures identified by Audit Analytics with the sample of ICD disclosures provided by Doyle et al. (2007a). We supplemented our U.S. sample by ICD disclosures collected from form 8-Ks by Doyle et al. (2007a). 12

15 advantage of the market-based approach is that there is no need to directly calculate the unobservable earnings quality; instead, we infer the earnings quality from the stock market s valuation of earnings. The disadvantage is that the market-based measure of earnings quality requires the assumption of market efficiency The association between the Section 302 ICD disclosure and Dechow and Dichev s (2002) earnings quality measure To construct Dechow and Dichev s (2002) earnings quality measure, we estimate the following cross-sectional regression within each industry-year (using Fama and French (1997) 48-industry classification) over the period from 1996 to 2002: WC t = β 0 + β 1 CFO t-1 + β 2 CFO t + β 3 CFO t+1 + β 4 REV t + β 5 PPE t + ε t (1) Where ΔWC is working capital accruals based on the Cash Flows Statement and is defined as (increase in accounts receivable + increase in inventory + decrease in accounts payable + decrease in income tax payable + net change in other accrued liabilities) scaled by average total assets. CFO is cash flows from operations scaled by average total assets. ΔREV is change in sales scaled by average total assets. PPE is gross property, plant, and equipment scaled by average total assets. Following McNichols (2002) and Francis et al. (2005a), we augment the original model in Dechow and Dichev (2002) by adding ΔREV and PPE to further capture the influence of economic fundamentals on accruals. The estimation period covers because a firm s ICDs may have existed for a while before the disclosure (Ashbaugh et al., 2007; Doyle et al., 2007b). We require at least three observations per firm and at least 20 observations per industryyear in the estimation. Each firm s earnings quality is measured as the standard deviation of the 13

16 residuals from the above cross-sectional model (denoted EQ_CROSS). Higher values of EQ_CROSS represent lower earnings quality. The estimation of EQ_CROSS relies on the assumption that the regression coefficients are the same for both U.S. firms and cross-listed firms within the same industry-year. To check the sensitivity of our results to this assumption, we also estimate a firm-specific regression by allowing the regression coefficients to vary across firms. Specifically, we estimate the following firm-specific time-series regression over the period , requiring at least 6 years of data per firm: WC t = β 0 + β 1 CFO t-1 + β 2 CFO t + β 3 CFO t+1 + ε t (2) To retain as many firms as possible, we do not include ΔREV and PPE in model (2). Each firm s earnings quality is estimated as the standard deviation of the residuals from the above time-series model (denoted EQ_TIME). Similar to EQ_CROSS, higher values of EQ_TIME indicate lower earnings quality. In subsequent empirical analyses, both earnings quality measures are expressed in natural logarithm to reduce the skewness. Prior literature has suggested several alternative accounting-based measures of earnings quality such as abnormal accruals or earnings smoothing. We choose the Dechow and Dichev (2002) measure as our primary proxy for several reasons. First, the Dechow and Dichev measure is the primary earnings quality proxy used by Doyle et al. (2007a) and Ashbaugh-Skaife et al. (2008a), the two studies that we benchmark with. Second, as Doyle et al. (2007a) note, ICDs will lead to both intentional and unintentional estimation errors in reported earnings. Because EQ_CROSS and EQ_TIME are designed to capture both types of estimation errors, they are preferred to other earnings quality proxies such as abnormal accruals and earnings smoothing 14

17 that more closely relate to managerial manipulation of earnings. Nevertheless, as a robustness check, we examine several alternative accounting-based earnings quality proxies in Section 7. To test the association between the Section 302 ICD disclosure and Dechow and Dichev s earnings quality proxy across cross-listed firms and U.S. firms, we estimate the following cross-sectional model modified from Doyle et al. (2007a): ln(eq) = β 0 + β 1 ICD + β 2 FOREIGN + β 3 ICD FOREIGN 12 + β n Controls + β n Controls FOREIGN + ε (3) n= 4 21 n= 13 The dependent variable, ln(eq), is the logarithm transformation of EQ_CROSS or EQ_TIME estimated using the Dechow and Dichev (2002) methodology. Among the independent variables, ICD is an indicator variable that equals one if the Section 302 disclosure reports an ICD (material weakness disclosure for reported results) from September 2002 to July 2006, and zero if there is no ICD disclosed during the same period. FOREIGN is an indicator variable that equals one for cross-listed firms, and zero for U.S.-domiciled firms. Controls refers to a vector of control variables discussed below. The coefficient on ICD captures the difference in earnings quality between U.S. ICD disclosers and U.S. non-icd disclosers. The coefficient on ICD FOREIG represents the incremental effect of ICD for cross-listed firms. 10 If both U.S. firms and cross-listed firms exert reasonable effort to detect and truthfully report ICDs, we should expect the ICD disclosure to be equally informative about earnings quality and therefore the coefficient on ICD should be 10 Although existing earnings quality proxies (including EQ_CROSS and EQ_TIME) are noisy and even biased, we have no reason to believe that our variable of interest ICD FOREIGN will be affected because any noises or biases would apply to both U.S. firms and cross-listed firms. In addition, we analyze regression model (3) for cross-listed firms domiciled in weak versus strong investor protection countries separately. 15

18 positive while the coefficient on ICD FOREIGN should be zero. 11 Ashbaugh-Skaife et al. (2008a) and Doyle et al. (2007a) have shown that U.S. firms Section 302 ICD disclosure is informative about earnings quality. As a result, we expect β 1 > 0. As we have argued in the Introduction, if cross-listed firms have lower incentive to detect and truthfully report ICDs than U.S. firms, the coefficient on ICD FOREIGN is expected to be significantly negative, i.e., β 3 < 0. The sum of the coefficients on ICD and ICD FOREIGN captures the difference in earnings quality between cross-listed ICD disclosers and cross-listed non-icd disclosers. If cross-listed firms Section 302 internal control disclosures are not informative at all about earnings quality, the sum of the coefficients on ICD and ICD FOREIGN should be statistically insignificant, i.e., (β 1 + β 3 ) = 0. Following Doyle et al. (2007a), we include several control variables that potentially relate to earnings quality and firms ICD disclosure decision. MEAN_%LOSS is the ratio of the number of years of losses relative to the total number of years of data from SALES_VOLATILITY is the standard deviation of the ratio of annual sales to average total assets over CFO_VOLATILITY is the standard deviation of the ratio of operating cash flows to average total assets over MEAN_ASSETS is the average total assets over MEAN_CYCLE is the average operating cycle, defined as [average accounts receivable/(sales/360) + average inventory/(cost of goods sold/360)] over MEAN_SEGMENTS is the average number of total operating and geographic segments over FIRM_AGE is the number of years the firm has CRSP data as of EXTREME_GROWTH is an indicator variable equal to one if the average industry adjusted sales 11 Our maintained assumption is that the impact of an ICD of the same type on earnings quality is the same across U.S. firms and cross-listed firms. Section discusses how the violation of this assumption may affect the interpretation of our results. 16

19 growth over is in the top quintile, and zero otherwise. RESTRUCTURE_CHARGE is the magnitude of restructuring charges scaled by market value of equity over We measure the control variables over the time period to be consistent with the measurement of EQ_CROSS and EQ_TIME. To reduce multicollinearity, we standardize all the continuous independent variables. Except for MEAN_%LOSS, EXTREME_GROWTH, and RESTRUCTURE_CHARGE, all the other control variables are converted into natural logarithm in estimating regressions to reduce the skewness in the data. Similar to Doyle et al. (2007a), we predict that ln(eq) increases with MEAN_%LOSS, SALES_VOLATILITY, CFO_VOLATILITY, and MEAN_CYCLE, and decreases with MEAN_ASSETS The association between the Section 302 ICD disclosure and the earnings response coefficient Our second approach uses the earnings response coefficient to examine the usefulness of the Section 302 ICD disclosure in assessing earnings quality (referred to as the market-based approach). Following Easton and Harris (1991), we regress the annual abnormal stock returns (cumulated over the 15-month period ending three months after the fiscal year-end) on the level and the change of annual earnings (defined as earnings before extraordinary items scaled by the beginning-of-year market value of common equity). The sum of the coefficients on the level and the change of earnings indicates the informativeness of earnings. Earnings informativeness has been widely used as a market-based measure of earnings quality (e.g., Fan and Wong, 2002; Francis et al., 2005b; Warfield et al., 1995; Wang, 2006). Greater earnings informativeness implies higher earnings quality. 12 The regression results of model (3) are robust to the inclusion of Fama and French s (1997) industry dummies and their interactions with FOREIGN (each industry is required to have a minimum of 100 observations to preserve the degrees of freedom) or an auditor quality variable (BIG4, an indicator variable that equals one when the firm s auditor is one of the Big 4 auditors, and zero otherwise) and its interaction with FOREIGN. The coefficients on the latter two variables are mostly insignificant. 17

20 To examine the association between the Section 302 ICD disclosure and the earnings response coefficient, we estimate the following model across cross-listed firms and U.S. firms over the period : CAR it = β 0 + β 1 ICD + β 2 FOREIGN + β 3 ICD FOREIGN + β 4 E it + β 5 E it ICD + β 6 E it FOREIGN + β 7 E it ICD FOREIGN + β 8 E it + β 9 E it ICD + β 10 E it FOREIGN + β 11 E it ICD FOREIGN 20 + β n Controls + β n E it Controls + β n Controls FOREIGN n= n= n= 30 + β ne it Controlsr FOREIGN + β n E it Controls n= n= 48 + β n E it Controls FOREIGN + ε it (4) n= 57 The dependent variable, CAR, is cumulative market-adjusted returns for the 15-month period ending three months after the fiscal year-end. 13 E and ΔE are the level and the change of earnings before extraordinary items scaled by year-beginning market value of equity. The sum of the coefficients on E ICD and ΔE ICD, or (E+ΔE) ICD, represents the difference in earnings informativeness between U.S. ICD disclosers and U.S. non-icd disclosers. To the extent that U.S. firms Section 302 ICD disclosure indicates poor earnings quality, we expect the sum of these coefficients to be significantly negative, i.e., (β 5 + β 9 ) < 0. If cross-listed firms ICD disclosure is less informative about earnings quality than U.S. firms ICD disclosure, the sum of the coefficients on E ICD FOREIGN and ΔE ICD FOREIGN, or (E+ΔE) ICD FOREIGN, is predicted to be significantly positive, i.e., (β 7 + β 11 ) > 0. If crosslisted firms internal control disclosure is not informative about earnings quality, we expect the sum of the coefficients on the term (E+ΔE) (ICD+ICD FOREIGN) to be statistically insignificant, i.e., (β 5 + β 9 + β 7 + β 11 ) = As cross-listed firms have up to 6 months after the fiscal year-end to file their 20-F, we also run a sensitivity check using an 18-month CAR that ends six months after the fiscal year-end and obtain similar results. 18

21 Following Ghosh and Moon (2005) and Ghosh and Lubberink (2006), we also include the following control variables in estimating model (4). BM is book-to-market ratio. LEV is book leverage ratio defined as total debt divided by total assets. ASSETS is total assets. All three variables are measured at the end of the prior year. PERSISTENCE is the earnings persistence coefficient from a firm-specific regression of year t+1 earnings scaled by the average total assets on year t earnings scaled by the average total assets using the past 5 years data. VOLATILITY is the standard deviation of earnings scaled by the average total assets using the past 5 years data. BETA is the market beta computed using the past 36 monthly stock returns. BIG4 is an indicator variable that equals one when the firm s auditor is one of the Big 4 auditors, and zero otherwise. REGULATION is an indicator variable that equals one for regulated firms (two-digit SIC code between 40 and 49 or between 60 and 63), and zero otherwise. FIRMAGE is the number of years that the firm is publicly listed in the U.S. as of the fiscal year-end. To reduce multicollinearity, we standardize all the continuous independent variables. 5. Empirical results on the association between the Section 302 ICD disclosure and earnings quality 5.1. Descriptive statistics Table 1 reports the descriptive statistics for the sample. Panel A (Panel B) reports the descriptive statistics for U.S. firms and cross-listed firms with non-missing EQ_CROSS (EQ_TIME). The descriptive statistics for cross-listed firms are shown for strong and weak investor protection countries separately. For brevity, we do not report the descriptive statistics for the control variables in Panel B since they are similar to those reported in Panel A. As shown in Panel A, the percentage of firms reporting internal control deficiencies (ICD) or material weaknesses (MW) monotonically increases from U.S. firms to cross-listed 19

22 firms domiciled in strong investor protection countries to cross-listed firms domiciled in weak investor protection countries, although the differences are insignificant except for the difference in ICD between U.S. firms and cross-listed firms from weak investor protection countries (twotail p-value = 0.056). One may conclude from these simple frequency counts that there is no evidence that cross-listed firms are more likely to conceal existing ICDs than U.S. firms. Unfortunately, this is an erroneous conclusion because we do not know the true frequencies of the ICDs for U.S. firms and cross-listed firms. To illustrate, suppose that the true percentage of firms with ICDs is higher for cross-listed firms than for U.S. firms (say 50% versus 30%). As argued in the Introduction, cross-listed firms are less likely to report ICDs than U.S. firms. As a result, let s assume that the percentage of firms that actually report ICDs is 30% for both crosslisted firms and U.S. firms. Thus, relative to the true frequencies of ICDs, cross-listed firms conceal existing ICDs to a greater extent than U.S. firms since a portion of the cross-listed firms that have ICD problems (i.e., 20%) fail to disclose the ICDs. However, by observing the identical frequencies of reported ICDs (i.e., 30%), one would erroneously conclude that cross-listed firms report ICDs as truthfully as U.S. firms. Panel A also shows that EQ_CROSS is similar across the two subsamples of cross-listed firms but is significantly higher for U.S. firms, while Panel B reports that EQ_TIME is statistically similar across U.S. firms and cross-listed firms domiciled in weak investor protection countries. 14 Although prior research has shown that better earnings quality is attributable to less severe agency problem (e.g., Klein, 2002), one cannot infer from the above descriptive statistics that cross-listed firms agency problem is less severe than U.S. firms because the two samples differ in many other dimensions. For example, relative to cross-listed firms, sales volatility 14 This result seems inconsistent with Lang et al. (2006). However, due to many differences between our study and Lang et al. (2006) (e.g., sample size, sample selection criteria, earnings quality measurement, and research design choices), it is difficult to reconcile the difference between our result and Land et al. s. 20

23 (SALES_VOLATILITY) and cash flow volatility (CFO_VOLATILITY) are significantly higher and the average assets (MEAN_ASSETS) are significantly lower for U.S. firms. Prior research has shown that EQ_CROSS and EQ_TIME increase with SALES_VOLATILITY and CFO_VOLATILITY and decrease with MEAN_ASSETS (Dechow and Dichev, 2002; Doyle et al., 2007b; Ashbaugh-Skaife et al., 2007) Results on the association between the Section 302 ICD disclosure and Dechow and Dichev s (2002) earnings quality measure Table 2 reports the regression results of model (3). Columns (1) and (4) provide results based on all cross-listed firms and U.S. firms. The results for the other columns will be discussed in Section 6.1. Note that the results in Table 2, as well as those in Tables 3 7, consider material weakness disclosures only but the inferences are similar if we include all three types of ICDs. As shown in Table 2, the coefficients on ICD are significantly positive at the 0.05 level or lower using either EQ_CROSS or EQ_TIME as the dependent variable. This finding is consistent with recent evidence (Doyle et al., 2007a; Ashbaugh-Skaife et al., 2008a), suggesting that U.S. firms Section 302 ICD disclosure contains useful information about earnings quality. However, the coefficient on ICD FOREIGN is significantly negative, suggesting that cross-listed firms' ICD disclosure contains significantly less information about earnings quality than U.S. firms'. In addition, we find the sum of the coefficients on ICD and ICD FOREIGN is statistically insignificant using either EQ_CROSS or EQ_TIME. Thus, in sharp contrast to U.S. firms, crosslisted firms Section 302 ICD disclosure is statistically unrelated to earnings quality. 15 Given the significant difference in firm size between U.S. firms and cross-listed firms, we also estimate model (3) after deleting small U.S. firms to equalize the median MEAN_ASSETS between the remaining (relatively large) U.S. firms and cross-listed firms. Our results remain qualitatively similar. 21

24 Turning to the control variables, our results are generally consistent with those reported in Doyle et al. (2007a, Table 2). Although the coefficients on some of the control variables are significantly different between U.S. firms and cross-listed firms, overall these control variables explain EQ_CROSS and EQ_TIME equally well for both samples of firms Results on the association between the Section 302 ICD disclosure and the earnings response coefficient We now turn to the association between the Section 302 ICD disclosure and the earnings response coefficient as described in model (4). To be consistent with the estimation of model (3), we estimate model (4) over the same period from 1996 to As the Section 302 ICD disclosure started in 2002, we implicitly assume that investors can infer a firm s ICDs through alternative information sources prior to This assumption is plausible because most economic determinants of ICDs are publicly known (e.g., see Ashbaugh-Skaife et al., 2007). Indeed, Ghosh and Lubberink (2006) find that investors anticipated at least a portion of the ICDs prior to the first disclosure of ICDs. As a sensitivity check, we also estimate model (4) over the following three alternative time periods and obtain similar results: 1) use all the years from 1996 up to the year of the first Section 302 ICD disclosure for ICD disclosers and for non- ICD disclosers; 2) use all the years from 1996 up to the year of the first Section 302 ICD disclosure for ICD disclosers and for non-icd disclosers; and 3) use all the years from 2002 to 2006 for both ICD disclosers and non-icd disclosers. Table 3 reports regression results of model (4). Column (1) provides result based on all cross-listed firms and U.S. firms. The results for the other columns will be discussed in Section 6.1. As shown, the coefficient on the term (E+ΔE) ICD is significantly negative, consistent with the notion that U.S. firms Section 302 ICD disclosure provides useful information about 22

25 earnings quality. As conjectured, the coefficient on the term (E+ΔE) ICD FOREIGN is positive, although statistically insignificant. Moreover, the coefficient on the term (E+ΔE) (ICD+ICD FOREIGN) is statistically insignificant, confirming the inference drawn from Section 5.2 that cross-listed firms Section 302 ICD disclosure is not informative about earnings quality. Turning to the control variables, we find that non-regulated firms, young firms, and firms with higher growth potential, lower financial leverage, larger market value, and less volatile earnings are associated with higher earnings response coefficients (ERC). The ERC coefficients for PERSISTENCE and BETA are also consistent with our expectations but statistically insignificant. The ERC coefficient for BIG4 is negative. This result is contrary to the intuition but nevertheless consistent with Ghosh and Moon (2005) and Ghosh and Lubberink (2006). 6. Identifying the causes for the weaker association between the Section 302 ICD disclosure and earnings quality for cross-listed firms There are potentially two non-mutually exclusive explanations for the weaker association between the ICD disclosure and earnings quality for cross-listed firms. First, cross-listed firms weaker results are due to their management s lack of incentive to expend reasonable effort to detect and truthfully disclose existing ICDs. In Sections 6.1 and 6.2 we provide two pieces of empirical evidence that are consistent with this explanation. Second, the ICD disclosure is equally informative about earnings quality for both cross-listed firms and U.S. firms but crosslisted firms weaker results are due to some limitations of either our data sources or research designs. We consider this alternative explanation in Section